Ran into an interesting problem while building EchoAide. If a web socket connection drops mid-consultation, how would you resume streaming without losing audio? My thought: persistent recordingSessionId + grace window + replay buffered chunks after reconnect. Thoughts?
Quick update on EchoAide’s Architecture -
EchoAide: browser audio → WebSocket → Soniox (live STT) → NestJS → LangGraph → structured note in Postgres.
Frontend listens via Supabase Realtime on the note ID.
Open to any feedback!
@JacksonnCreator Hey, I'm Aryan Nandanwar, a Software Engineer currently building EchoAide, an AI copilot for outpatient clinics that listens to doctor-patient conversations and generates structured clinical notes.
https://t.co/gXY6grjBjL
@phoenix1454 The latency does take a hit with multiple LLM nodes. Right now it’s just a single LLM node for note generation
. I’m using LangGraph because the workflow will eventually branch into things like insurance coding, lab orders, prescriptions, and follow-ups.
@orynth I’ve built EchoAide - an AI copilot for outpatient clinics that listens to doctor-patient conversations and generates structured clinical notes.
https://t.co/gXY6grk99j
@yaya_bn8 Hey Daniel, I've built EchoAide, an AI SaaS platform for outpatient clinics that handles patient intake, AI-powered clinical documentation, and follow-ups.
https://t.co/QJ1ziZUiqP
https://t.co/2iqmdroWyn
https://t.co/ZFP5svmfKs